小波MFCC和HMM在低空声目标识别中的应用

Application of wavelet MFCC and HMM in low altitude acoustic target recognition

摘要:

低空目标声信号是时变的、非平稳的，若以传统的Mel倒谱系数(MFCC)作为特征参数，虽然能模拟人耳的听觉特性，但不能更好地反映低空声信号的动态特性。基于小波变换可以精确反映非平稳信号瞬间变化的特点，文中用小波变换替代MFCC参数提取过程中的FFT变换，将改进后的MFCC与能够很好表征时变信号特性的隐马尔可夫模型(HMM)结合，对低空目标声信号进行检测识别。最后用实际采集到的直升机声信号和非直升机声信号进行实验，取得了较好的识别效果。实验的对比结果验证了该方法的有效性和可行性。

Abstract:

Although Traditional Mel Frequency Cepstrum Coefficient(MFCC) can simulate human auditory characteristics, it can not reflect the dynamic characteristics of low altitude target acoustic signal satisfactorily because of its time-varying and non-stationary properties. Wavelet transform which can exactly reflect instantaneous change of non-steady signal can be used to replace the FFT transform in extracting MFCC parameter process. And then the improved MFCC, combined with Hidden Markov Model(HMM) which can behave better in representing the time-varying signal, is used to recognize the low altitude acoustic target. Experiments with actually gathered helicopter acoustic signal and non-helicopter acoustic signal are performed, and better recognition effect is obtained. The validity and feasibility of the proposed detecting and recognizing algorithm are verified by the experimental comparison results.